Use of Vehicle Probe and Cellular GPS Data by State Departments of Transportation (2021)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

5 Probe and Cellular GPS Data Collection Methods Vehicle location information can be collected or extracted in several different ways: • In-vehicle embedded devices. • GPS devices embedded in cell phones. • Cell phone service providers’ networks. • Cell phone applications. In-Vehicle Embedded Devices In-vehicle embedded devices are part of vehicle manufacturer-built systems that collect vehicle telematics data including vehicle location, vehicle performance, and driver behavior. Unlike handheld devices, vehicles are not restrained by the small form factor and therefore necessary components can be distributed throughout the vehicle, while leveraging a vehicle’s power system for operation and vehicle velocity information for enhancing location information (Jagoe 2003). Location information is captured by fusing in-vehicle GPS receiver information and Dead Reckoning (DR) generated information. The DR method uses a previously determined position and the vehicle’s velocity to approximate the vehicle’s location (Cho and Choi 2006). GPS Devices Embedded in Cell Phones Most cell phone devices are equipped with a GPS receiver that can obtain navigation messages from at least three satellites to determine the device’s latitude and longitude (and altitude if using a fourth satellite signal). In addition to receiving a GPS signal, many devices utilize a method called assisted-GPS where wireless network information is used to supplement GPS information. This approach reduces the power consumed by the headset, optimizing start-up and acquisition time, and increasing the sensitivity of the GPS device (Barnes 2003). Cell Phone Service Providers’ Networks Cell phone service providers have the capability of tracking the location of individual devices as they move through the network. There are several network-level tracking methods generally used to locate a cell phone device. The Cell of Origin method approximates a user’s location based on the location of the base station serving the mobile device. This method has low accuracy of 150 to 10,000 meters based on individual cell size in the network (Barnes 2003). This method is supplemented by timing C H A P T E R 2 Literature Review

6 Use of Vehicle Probe and Cellular GPS Data by State Departments of Transportation advance information which calculates the time difference between the start of a radio frame and a data burst, which is used to further approximate a user’s location within a cell (Jagoe 2003). The Estimated Time of Arrival and Angle of Arrival methods use the time necessary for a device’s signal to reach multiple base stations and the direction from which the signal arrives, respectively. This information is then sent to a mobile switch where it is analyzed to generate the approximate location of the mobile device (Barnes 2003, Jagoe 2003). Cell Phone Applications Cell phone applications or apps often utilize device location (collected using a device’s GPS receiver and network information) to enhance user experience. When users install social media, weather, or similar apps that use location information for other purposes, their location informa- tion is captured and used to provide additional services and often resold to third parties (Huang et al. 2018). History of Probe Data for Transportation Applications Probe-Based Speed Data In the early 2000s, transportation agencies identified a need to supplement data from the traditional static sensor technology to better evaluate real-time and historical performance of the transportation system. Early efforts included generating floating car data using a single vehicle equipped with a GPS unit traversing corridors at specific times. The resulting GPS points were snapped to an underlying map to generate metrics such as average speed, travel time, and congestion index (Tong et al. 2005). The use of this floating car method was intermittent, tedious, and time-consuming, and the method proved to be only marginally useful (Pack et al. 2019). Radio camera positioning of cellular phones was also being developed at this same time. This technology used the signal strength of the cell phone at the antenna site and various forms of triangulation to identify the position and speeds of vehicles for use in traveler information and traffic control (Smith et al. 2001). Other methods relied on the cooperation of cellular carriers to mine the signal timing and handoff data emanating from a cellular tower-switching network (Myr 2003). Some agencies have been generating their own probe vehicle data using toll tags. For example, the Florida DOT (FDOT), the Texas DOT, and toll authorities around the country use toll tags to identify and re-identify vehicles as they traverse the toll facility and use that information to calculate speed and travel time between re-identification points (Pack et al. 2019). In 2006, new vehicle probe technology was emerging as a means of continuously monitoring traffic as commercial firms were offering traffic data services based on a variety of methods, the most common of which were based on either cellular geo-location or fleet GPS-based telematics reporting technologies. (Center for Advanced Transportation Technology, University of Maryland and KMJ Consulting Inc. 2011). The use of probe vehicle data for real time operations and planning among agencies was expanded to a larger number of agencies when The Eastern Transportation Coalition (formerly the I-95 Corridor Coalition) created a probe vehicle data marketplace in 2007 (University of Maryland 2007). This request for proposal (RFP) outlined desired requirements for both data and quality that propelled use of probe vehicle data among coalition member agencies and later other agencies across the country. It also resulted in a model data use agreement that has been cited in many publications and RFPs in the years following this initial RFP.

Literature Review 7 Commercial speed data providers have moved away from using a single method, technology, or source to generate probe data, to combining probe data from multiple sources and technolo- gies to create a comprehensive traffic information service (Center for Advanced Transportation Technology, University of Maryland and KMJ Consulting Inc. 2011). These offerings have largely replaced the need for fixed roadside infrastructure to support speed data collection. Origin-Destination Data O-D studies have been a core activity in the world of transportation to better understand travel patterns, identify traffic congestion sources and potential traffic control strategies, better plan transit services, and plan urban development for almost 100 years (McClintock 1927, Mickle 1944, Braff 1948, Blucher 1950). Traditionally, these studies relied on data collected through spotter observations, as well as paper-based and web surveys of travelers. As a result, they repre- sented a small sample of the population and required significant time and effort to administer and process, often producing results many years after commencing the study (Williams 1986, Hartgen 1992, Richardson 2003). For example, even the surveys themselves could be time-consuming because to accurately capture demand variability, multi-day surveys and, more specifically, a 2-week survey duration are necessary (Senbil and Kitamura 2009). Vehicle Telematics-Based Origin-Destination Data With the emergence of probe speed data in the mid-2000s, commercial firms such as AirSage, HERE, and INRIX began capturing O-D data and delivering that data as a product to agencies (Allos et al. 2014). The cost and latency of these products were lower than those of traditional surveys, and sampling rates were higher. Location-Based Services Data In the late 1990s and early 2000s, telecommunications companies, application developers, and content providers were looking to leverage the emerging LBS capabilities aggregated from multiple sensors in mobile devices (Myllymaki and Edlund 2002). Additional technology and algorithm advancement led to more accurate and easier ways to locate mobile devices and track device paths (Barbeau et al. 2008). These developments led to further refinements of the LBS, and produced data to not only identify locations and paths of users and their devices, but also to infer the mode of transportation to differentiate between cars, buses, and other modes (Byon, Abdulhai, and Shalaby 2009). In recent years, the proliferation of mobile apps (like social media apps and weather apps) that capture device location information enhanced the volume of available location data. For exam- ple, location-based social networks enable uploading geotagged content (e.g., photos, videos, and recorded routes), sharing the present location (e.g., “check-in” at Foursquare), commenting on an event at the place where it is happening (e.g., via Twitter), or leaving ratings/tips/reviews for a location (e.g., a restaurant) (Huang et al. 2018). Today, location information extracted from social media and other mobile device applications allows agencies and the private sector to analyze a variety of transportation problems, such as parking supply and demand implications (Mondschein et al. 2020), trip purpose, the reliability of bike share programs (Svartzman et al. 2020), communications during natural disasters affecting the transportation system (Lovari and Bowen 2020, Roy et al. 2020), the analysis of transit services and ride hailing platforms (Kim et al. 2019), safety and mobility performance in smart cities (Oh et al. 2019), and others. Agencies began getting access to O-D and trajectory data based on LBS. Third-party data providers such as StreetLight collect and aggregate LBS data and package it for use by the agencies (Lee and Sener 2017).

8 Use of Vehicle Probe and Cellular GPS Data by State Departments of Transportation Trajectory Data In the last several years, some of the vehicle telematics-based O-D data providers began adding value to the traditional O-D data sets by including associated trajectory data. The new products consist of origins and destinations, as well as waypoints captured between those origins and destinations (Petrone and Franz 2018, Fan et al. 2019). Combining O-D and trajec- tory information provides a comprehensive view of “trips” taken by individual vehicles. These trips provide valuable information previously unavailable, such as routes taken, time to traverse those routes, and variation in travel time and speed along the route. Most of these O-D and trajectory data offerings are sourced directly from vehicle telematics available from integrated sensors and devices from vehicle original equipment manufacturers and vehicle fleets (Sharma et al. 2017). Types and Properties of Current Probe and Cellular GPS Data Available The literature review covered three main classes of probe and cellular GPS data: speed, O-D, and trajectory (sourced from telematics and LBS data). Speed Data Probe-based speed data are derived from vehicles and mobile devices equipped with embedded GPS devices. Speeds and travel times provide real-time or historic average speeds on one or more road segments. To be useful, probe vehicle data must be conflated to a roadway network. Initial commercial offerings relied on Traffic Message Channel (TMC) codes, a reference system designed to give a unique alpha-numeric code to each road segment for purpose of assigning traffic information to that segment (HERE 2015). These TMC codes are typically assigned at significant decision points, interchanges or intersections in an unambiguous format, independent of map vendor. The North America Location Code Alliance created, owns, maintains, and expands the U.S./Canada TMC location code table that adheres to the international standard on location referencing (INRIX 2018). While being the only standardized coding method to uniquely identify roadway segments for the purpose of conveying traffic and other information, the TMC code segmentation reached its limitations in road coverage, the ability to cover new roads more quickly, segment overlap and gapping, and segment resolution, as probe data have become denser and more granular. INRIX introduced XD segments in 2013, a proprietary segmentation that addresses some of the TMC code limitations (Young et al. 2015, INRIX 2018). Similarly, other data providers are offering sub-TMC products to address TMC segmentation limitations (HERE 2015, TomTom 2015). The National Performance Management Research Data Set In July 2013, the Federal Highway Administration (FHWA) procured the NPMRDS to support the Freight Performance Measures and Urban Congestion Report programs, as well as the Moving Ahead for Progress in the 21st Century Act (MAP-21) performance management activ- ities. This data set consists of actual observed average travel times every 5 minutes, 24 hours a day, 7 days a week covering the National Highway System (NHS) as defined by MAP-21 and delivered monthly starting with October 2011. The average travel time data are sepa- rated for freight, passenger, and all traffic. The first version of the NPMRDS was provided by HERE Inc. and sourced from mobile phones, vehicles, and portable navigation devices for passenger vehicles, and from the American Transportation Research Institute’s leveraging of embedded fleet systems for freight vehicles (FHWA Office of Operations and Resource Center).

Literature Review 9 Since February 2017, data are provided by a team led by the University of Maryland Center for Advanced Transportation Technology (Center for Advanced Transportation Technology). The speed and travel time data are provided by INRIX, which leverages its existing source data, including millions of connected vehicles, trucks, and mobile devices that anonymously supply location and movement data. The data cover more than 400,000 segments provided in 5-minute intervals, 24 hours a day (NPMRDS FAQ 2020). Origin-Destination Data Probe-based O-D data contain basic information about trips between two geographic points. It does not contain information about specific routes between those two geographic endpoints. The geographic endpoints are reported as latitude/longitude pairs, but often generalized to zones. The O-D data set provides counts of trips between the origin zone and destination zone for a selected time period, such as workday a.m. peak period. The O-D data may also be broken down by travel mode and vehicle class type as well, such as passenger vehicles, heavy vehicles, medium vehicles, bikes, or pedestrians. (Pack et al. 2019, Southwest Washington Regional Transportation Council 2019, StreetLight 2020). Probe-based O-D data can be combined with traditionally available geodemographic infor- mation for origin and destination zones (Martin et al. 2018). Data may include standard census statistics as well as expanded data elements, such as population size and density, employment statistics, average income, and age, gender, race, and occupation statistics for a zone (Kim et al. 2012, Lawson 2018, StreetLight 2020). Trajectory Data Whereas O-D data provide two data points per trip, trajectory data provide information about the origin, destination, and routes taken between those endpoints. Trajectory data are timestamped location data from vehicles, cell phones, and other GPS-enabled devices throughout the network, often referred as “bread-crumb trail” data. Other data elements can include (Pack et al. 2019): • Unique device/vehicle ID, • Unique trip ID, • Departure time and location (trip origin), • Arrival time and location (trip destination), • Periodic “waypoints” during the trip, including latitude/longitude, where the period can be multiple times per second or once every X minutes, • Instantaneous speed/heading, • ID of road segment for the waypoint, and • Data that describe the path and routes of a trip from an origin to a destination. Trajectory data are available everywhere that probe speed data are collected, which means that it has ubiquitous coverage not otherwise available using traditional O-D surveying and floating car studies. The capture rate for trajectory data varies by region. For example, in Utah INRIX provides trajectory data that cover 3% of all trips in the state (Markovic et al. 2020). Trajectory data can be processed and made available within hours of collection at a higher price than that of data that are delivered weekly or monthly. Location-Based Services Data LBS use smartphones’ GPS technology (or control plane locating for older devices) to geolocate the device and integrate this location information with relevant services and content databases. The content databases provide supporting information such as the road network (digital maps),

10 Use of Vehicle Probe and Cellular GPS Data by State Departments of Transportation business and landmark information (points of interest), and dynamic data such as traffic and weather reports. The user’s past, present, or future location is joined with other context infor- mation to deliver a service (Schiller and Voisard 2004). For use in transportation, LBS data are collected when smartphone apps include code that collects location information from users that opt-in to use the app and report location informa- tion (Cuebiq 2019). The collected data are then packaged as a product that provides anonymous timestamped location information, including dwell times associated with points of interest (Cuebiq 2020). LBS data can then be further enhanced using inferred information such as income, race, education, and family status at the neighborhood level (StreetLight 2020). Current Primary Uses of Probe and Cellular GPS Data According to Pack et al. 2019, probe vehicle data conflated to underlying roadway network and in combination with other data sources can be used to support many different activities including but not limited to the following: Probe-Based Speed Data • Monitoring real-time congestion – Detecting and identifying incidents. – Issuing traveler information. – Conducting work zone monitoring and impact analysis activities. – Detecting the end of the queue. – Comparing real-time speed information to historical trends. – Identifying recurring and non-recurring bottlenecks. – Evaluating regional operations and situational awareness, for example, Metropolitan Area Transportation Operations Coordination (MATOC) (Harrison et al. 2019). • Performance management – Evaluating performance metrics over time: travel time, buffer time, reliability, planning time, and associated indices. – Incorporating data into dynamic performance management dashboards. – Investigating user delay cost. – Meeting legislative requirements, for example, MAP-21 and FAST Act target setting (Pu and Meese 2013, Vandervalk 2018). – Evaluating the worst bottlenecks in a region for a period of time. – Studying trends, including special event, holiday, and seasonal movements. – Exploring the impacts of capital investments prior to, during, and after completion of the project. – Conducting after action reviews, for example, Maryland State Highway Administration (Harrison et al. 2019). – Evaluating winter performance management, for example, the Ohio DOT (Harrison et al. 2019). – Measuring freight performance (Habtemichael et al. 2015). – Measuring truck and auto performance (Eshragh et al. 2015). • Planning and research – Identifying problems. – Prioritizing projects. – Performing safety analyses. – Implementing public participation/information campaigns. – Conducting before-and-after studies.

Literature Review 11 – Reporting project assessments, for example, the New Jersey DOT Tactical-level Asset Management Plan (Harrison et al. 2019). – Evaluating at-grade railroad crossings (Hafeez and Kasemsarn 2017). • Traveler information – Providing real-time travel time information on dynamic message signs (DMS). – Delivering network performance information. – Distributing special event and holiday guidance. Origin-Destination and/or Trajectory Data • Real-time traffic pattern analysis – Evaluating corridor demands based on observed vehicle trips. – Evaluating the effectiveness of implemented detours around incidents and congestion and observing self-detouring patterns. – Identifying temporary but significant changes in origin-destination patterns and route utilization resulting from special events or closures. • Signals performance and turning movement analysis (Center for Advanced Transportation Technology, INRIX 2020) – Observed distribution of speeds through intersections. – Arrival on green metrics. – Turning movement metrics. – Intersection travel times. – Approach speeds. – Intersection delays. • Multi-modal system utilization – Discovering mode transitions and influence traveler decisions based on network utilization patterns. – Analyzing freight patterns, for example, “Quantifying Long-Haul Trucks on Florida’s Highways” (StreetLight 2020). – Assessing the economic impact of mode utilization, for example, “Virginia Bike Tourism: Measuring Cycling’s Economic Impact” (StreetLight 2020). • Planning and research – Traditional origin-destination analysis to identify trip origins and destinations, work versus leisure travel, and many other data points. – Waypoint analysis to determine if traffic in a specific area (e.g., state, county, traffic analysis zone, or business center) originated in the same area, neighboring area, or another more distant location. This can identify whether certain corridors are mainly local travel or pass- through corridors. – Trip cluster analysis to evaluate the effectiveness of existing transit service or identifying areas in need of new transit service. – Analysis of trip patterns that may affect critical freight corridors or ports. – Infrastructure investment decisions, for example, analyzing metrics to know where to place electric vehicle charging station locations (StreetLight 2020). Data Quality, Accuracy, and Reliability Probe vehicle data are available from only a sample of total vehicles flowing through the network but are generally considered sufficient enough to estimate travel time distributions through the network. Early studies have shown that even a small percentage (1%-2%) can be adequate for producing data appropriate for certain congestion-related performance measures (Hunter et al. 2009).

12 Use of Vehicle Probe and Cellular GPS Data by State Departments of Transportation Early studies attempted to quantify the quality of probe-based speed data and its ability to be used for certain applications on various classes of roadways based on AADTs or signal spacing. However, the density, granularity, coverage, and quality of data has improved to where it has become a key source of information to support agency operations and planning efforts on most roads for most use cases (Pack et al. 2019). Extensive validation efforts have shown that probe vehicle data are accurate and sufficient enough to support most traffic operations and planning efforts (Vander Laan and Zahedian 2019-2020), as well as policy research (van der Loop et al. 2019). The U.S. DOT and many agencies have performed their own analysis of the NPMRDS, which is based on probe speed data, to support target setting and performance management (Turner and Koeneman 2018, Refai et al. 2017). Validation efforts for LBS and trajectory-related data are still in the early stages, but initial independent validation efforts have shown that the data are extremely representative of actual conditions. For example, multiple aerial photo studies have validated INRIX Trips data (Jordan et al. 2016a, 2016b, 2017), and the Maryland Transportation Institute’s methodologies for computing trips by mode from LBS data have also been validated (Zhang et al. 2020). Gaps in Data Usage Information Many agencies consider travel time and probe-generated speed data as the top three data elements they would be interested in acquiring if available (Pack and Ivanov 2014). However, prior research has shown that agencies struggle to work with the data due to the size of the data. Data for even a small state like Rhode Island can include hundreds of millions of trajectory records and billions of waypoints. It is for these reasons that third-party analytics packages are starting to be adopted by agencies.